Machine Learning Projects for Mobile Applications

TensorFlow-to-Core ML conversion

TensorFlow-to-CoreML conversion is done using the tf-coreml library. 

Check out the TensorFlow-to-Core ML convertor library at the following link: https://github.com/tf-coreml/tf-coreml. 

We have a three steps process for converting a TensorFlow model to Core ML.

1. The model we generated (output_graph.pb) in the last step requires support for power. Apple's Core ML tools provide a unary conversion which supports power. We need to add this code into the TensorFlow implementation as follows:

# tfcoreml src # file1 : _interpret_shapes.py # in the _SHAPE_TRANSLATOR_REGISTRY we need to add the Pow operation _SHAPE_TRANSLATOR_REGISTRY = { ... previous keys ... # add this: 'Pow': _identity, } # file 2: _ops_to_layers.py ...